Early Prediction of Heart Failure From Routine Cardiac CT Using Radiomic Phenotyping of Epicardial Fat

Epicardial adipose tissue (EAT) is a metabolically active visceral fat depot that is both a sensor and a modulator of myocardial biology and changes its composition in response to paracrine signals from the myocardium. We hypothesized that radiomic characterization of EAT from routine coronary computed tomographic angiography (CCTA) can noninvasively capture this adverse remodeling and enable early heart failure (HF) risk stratification.

We sought to develop and externally validate a reproducible radiomic signature of EAT associated with incident HF.

We conducted a multicenter cohort study of 72,751 adults without known HF or myocardial infarction undergoing CCTA across 9 UK centers (2007-2022). We deployed a fully automated pipeline to segment EAT and extract 1,655 volumetric, shape, and higher-order radiomic texture features. Using a harmonized survival autoencoder architecture, we derived the fat radiomic profile for HF (FRP_HF). The model was developed in 59,327 individuals from 7 centers (age 57 ± 13 years, 47.5% female) and externally tested in 13,424 participants from 2 geographically distinct centers (58 ± 12 years, 49.4% female). Survival models were adjusted for age, sex, and conventional risk factors, including coronary artery disease (CAD) severity and EAT volume.

Over a median follow-up of 5.1 and 4.0 years, 1,737 (2.9%) and 363 (2.7%) participants developed HF in the internal and external validation cohorts, respectively. FRP_HF demonstrated robust discrimination (C-statistics: 0.869 [95% CI: 0.850-0.889] internal; 0.850 [95% CI: 0.831-0.870] external). Each 25-percentile increase in FRP_HF was associated with a nearly 4-fold higher adjusted HF risk (adjusted HRs: 3.90 [95% CI: 3.13-4.84] internal; 3.79 [95% CI: 3.01-4.76] external; both P < 0.001), with individuals in the highest decile exhibiting a nearly 20-fold higher risk compared with the lowest decile. In the external cohort, addition of FRP_HF to conventional risk models, including EAT volume and CAD severity, significantly improved 5-year discrimination (ΔAUC: 0.047; 95% CI: 0.029-0.065) and net reclassification (NRI: 0.39; 95% CI: 0.29-0.48) and suggested net clinical benefit on decision curve analysis. The associations were consistent across demographic subgroups and across the ejection fraction spectrum.

Automated radiomic phenotyping of EAT from routine CCTA enables scalable, biologically informed stratification of future HF risk before clinical onset, positioning opportunistic imaging-based visceral fat profiling as a potential tool for precision prevention.